namesUnavailableShard

Whether a failure is the cluster being unreachable rather than a bad request or a broken server.

Two shapes, because Qdrant reports it two ways. Sometimes it names the shard or the replica. More often, when a peer is simply gone, it reports the fan-out: "1 of 1 read operations failed", with the reason underneath. Nothing in that second message contains the word shard, which is why a matcher keyed only on it read a dead node as a generic server error.

Both shapes require two halves, so an ordinary failure that happens to say "failed" is not read as a cluster diagnosis.